A global pattern for influenza activity

A global pattern for influenza activity

International Congress Series 1219 (2001) 87 – 94 A global pattern for inf luenza activity T.A. Reichert a,b,*, A. Sharmaa, S. Pardoa a Becton Dicki...

118KB Sizes 2 Downloads 31 Views

International Congress Series 1219 (2001) 87 – 94

A global pattern for inf luenza activity T.A. Reichert a,b,*, A. Sharmaa, S. Pardoa a

Becton Dickinson and Company, New Jersey, USA b Entropy Limited, Upper Saddle River, NJ, USA

Abstract In two papers at this conference, we showed influenza to be the cause of  70% of seasonal variation in human mortality. Seasonal mortality, therefore, is a surrogate for the force of influenza infection. We asked whether there might be a geographical pattern. Methods: Monthly all-cause mortality data were collected for 37 countries for 1962 – 1965, 1968 – 1971 and 1978 – 1981. Deaths above baseline of summer (DABS) were determined for three winters in each period, and the ratio to the summer baseline (SB) was calculated. The average of this ratio for each period was plotted against geographical factors. Spline fits were made of each plot. Results: Plots of average DABS/SB ratio versus latitude reveal an asymmetric bell shape. The outliers are instructive. Plots of DABS/SB against midwinter temperature are bell-shaped, also with instructive outliers. The height of both ‘‘bells’’ diminishes from 1960s to 1980s. Mortality curves of outlier countries, in the plot against temperature, exhibit a broadened winter peak. The peak of the curve lies between 5 and 9 °C and at 38 – 44° degrees latitude. Conclusions: A surrogate for force of influenza infection, seasonal mortality, varies smoothly with latitude and mid-winter temperature. The large variation argues that influenza control programs should be tailored locally. D 2001 Elsevier Science B.V. All rights reserved. Keywords: Seasonal mortality

1. Introduction Sakamoto-Momiyama [1] demonstrated that, as countries attain economic development, plots of monthly all-cause mortality change universally, from one which exhibits both a large summer peak and a broad and lower winter elevation, to one with a summer trough

* Corresponding author. 262 W. Saddle River Rd, Upper Saddle River, NJ 07458, USA. Tel.: +1-201-9349365; fax: +1-201-934-1467. E-mail address: doctom [email protected] (T.A. Reichert).

0531-5131/01/$ – see front matter D 2001 Elsevier Science B.V. All rights reserved. PII: S 0 5 3 1 - 5 1 3 1 ( 0 1 ) 0 0 3 8 2 - X


T.A. Reichert et al. / International Congress Series 1219 (2001) 87–94

and a sharp peak occurring in mid-winter. The summer peak likely represents gastrointestinal/diarrheal disease, which is eliminated by effective sanitation and water supply; and the winter peak is thought to result from an association with malnutrition and/or failures in food preservation late in winter. The former is usually resolved before the latter. Countries with developed economies have monthly mortality curves which are sharply peaked in a mid-winter month. This is true for both all-cause mortality and for diseasespecific mortality. In the paper by Reichert and Sharma [2] in these proceedings, such data are presented, and it is demonstrated that     

the winter-seasonal component of the mortality curve for each disease is the same, for all diseases, in each of three widely separated countries, seasonal variation differs by a factor of about two between the US and Japan, with Australia intermediate, seasonal mortality is most variable in the height of the mid-winter peak, and that variability is highly correspondent with country – local influenza activity, inhibition of local influenza activity (by comprehensive vaccination) dramatically reduced the seasonal component of mortality, and cessation of inhibition restored the pre-intervention level of seasonal variability.

We infer, then, that seasonal variation of disease-specific or all-cause monthly mortality is a good surrogate for influenza activity within a healthcare system. We asked whether the

Fig. 1a. The distribution of seasonal mortality as a percentage of total annual mortality for 37 countries versus latitude, for the time period 1962 – 1965. Three instructive outliers are called out.

T.A. Reichert et al. / International Congress Series 1219 (2001) 87–94


variability seen in the three countries might be a part of a global pattern for influenza activity.

2. Methods Monthly mortality data were obtained for approximately 50 countries for the periods 1962 –1965 [3], 1968 –1971 [4] and 1978 – 1981 [5]. Countries for which a plot of monthly mortality exhibited a summer peak or multiple, irregular peaks, were eliminated from the database; and countries for which data were not available for all periods were not considered. Thirty-seven countries in all three hemispheres met these criteria. The Summer Baseline for any winter period was defined as the average of the mortality for the preceding and following summer months (June – September, in the northern hemisphere, and December – March, in the southern hemisphere). Deaths above the baseline of summer (DABS) were defined as the sum of the non-negative differences between the monthly mortality and the summer baseline, for each non-summer month. The ratio of DABS to the total annual all-cause mortality is the fraction of deaths that occur in the winter season (DABSFrac). This fraction compensates, after a fashion, for differences among countries in population number and age distribution. The population centrum was defined, for each country, as a point placed at the approximate population-weighted center of gravity for that country. DABSFrac was

Fig. 1b. The distribution of seasonal mortality as a percentage of total annual mortality for 37 countries for the period 1978 – 1981. The same outliers appear.


T.A. Reichert et al. / International Congress Series 1219 (2001) 87–94

plotted for each country versus the latitude of the population centrum. The mean midwinter temperature (in January for countries in the northern hemisphere, July in the southern hemisphere) was obtained for capitol cities in each country [6]. These were also plotted against DABSFrac. The resulting scatter plots were fitted using the locally weighted regression (Lowess) [7].

3. Results Fig. 1a displays the distribution of DABSFrac with latitude for the earliest time period (  1965), and Fig. 1b, the latest period (  1980). There is significant scatter, but there is also an obvious bell-shaped relationship, which is somewhat lower in the later period. There are also obvious outliers, which appear consistently in all three periods and are noted in the figure. The USA, Canada and the British Isles all have in common the fact that their winter temperatures are atypical for their latitude. The first two are much colder, the last much warmer. Eliminating these outliers, Lowess fits for all periods are

Fig. 2. Locally weighted regression fits to the distribution of winter seasonal mortality as a percentage of total mortality versus latitude for 37 countries, for three time periods.

T.A. Reichert et al. / International Congress Series 1219 (2001) 87–94


Fig. 3. The seasonality of mortality as a percentage of annual mortality for 37 countries versus the mean temperature in a mid-winter month in the capitol city, averaged for the time period 1968 – 1971.

shown in Fig. 2. The result is a bell-shaped curve, which has a broad maximum spanning latitudes, 38 –43, for each period and a general lowering of the maximum with advancing time. DABSFrac was also plotted against mean, mid-winter month temperature. Fig. 3 displays this relationship, for the middle period (  1970). Again, a bell-shaped trend is obvious, with outliers, which are now different. Fig. 4 displays the Lowess fits for all three time periods.

4. Discussion The distribution among economically developed countries of the fraction of all-cause deaths, which occur as a winter excess above the summer baseline, varies smoothly and consistently in all three hemispheres, in both latitude and in mean temperature of a central mid-winter month. A locally weighted regression fit to these distributions makes the underlying pattern manifest as bell-shaped curves, with maxima between 38° and 44° latitude and at 5 – 9 °C. The bells are relatively symmetric, at least near the peak, dropping


T.A. Reichert et al. / International Congress Series 1219 (2001) 87–94

Fig. 4. Locally weighted regression fits to the distribution of winter seasonal mortality as a percentage of total annual mortality, versus the mean mid-winter temperature in the capitol city, for 37 countries, for three time periods.

to half of peak values at ± 12° of latitude and 10 °C. Data is relatively scant in low latitudes ( < 30 degrees) and high winter temperatures (>12 °C). Therefore, our ability to draw conclusions relative to tropical regions is limited. In latitude, the global pattern of seasonal mortality decreased with about equal amounts between each pair of the three time periods; about 25% in toto, in peak magnitude. This is consistent with the general worldwide reduction in the activity of influenza over the same time. Relative to mid-winter average temperature, the global pattern also declined, about 1/3 in magnitude, with less distinction between the two earlier periods. We are not the first to suggest a relationship between mid-winter temperature, seasonal mortality and influenza. Curwen [8], reviewing some of the same data, demonstrated that winter quarter mortality was strongly associated with both mean temperature and deaths attributed to influenza. Unfortunately, this author considered only monotonic relationships and, therefore, found data from the period around and after 1980 troubling to his hypothesis.

T.A. Reichert et al. / International Congress Series 1219 (2001) 87–94


The presently proposed pattern of influenza activity versus mid-winter mean temperature exhibits outliers not noted by previous authors. All such outlier countries (except Iceland, for which seasonal mortality is lower than trend) were members of the so-called Soviet bloc at the index time. The USSR and Bulgaria are consistently outliers in each time period. There are at least two plausible explanations. Either the temperature assigned is lower than a more appropriate average, or some consistent element acted to elevate the apparent seasonality. The additional outliers, in Fig. 3, suggest that one such element might be a broadening of the winter mortality peak. Each of the ‘‘outlier’’ countries, for that period, has a relatively broad winter peak, spanning 3 or 4 months, rather than 1 or 2. Many factors might conspire to produce such a broadening. However, SakamotoMomiyama [1] noted that the sharpening of the winter peak occurs relatively later in economic development than does the loss of the summer peak. Factors such as failure of food storage facilities, food scarcity, and in particular, lack of availability of vitamin-rich foodstuffs may produce both malnutrition and an increased susceptibility to an array of diseases. The above are plausible, but remain to be speculative bases for the noted departure from the apparent pattern. Seasonal variation in all-cause mortality is relatively easy to measure in any healthcare system, and these data appear to be reliable at low levels of sophistication in data collection. For this reason, the surrogate association of the winter excess of mortality with influenza activity is a powerful tool for both global assessment and pandemic planning. Figs. 2 and 4 readily demonstrate that influenza activity in severely affected regions may be large integer factors greater than that in other countries of similar levels of economic development. We note that countries located near the peak of activity are those which are often cited as the plausible origin of historical influenza pandemics, and cities with peak characteristics are overwhelmingly represented in the list of viral isolates used in vaccines in the modern era. The above model does not include factors plausibly important in the rise and propagation of influenza epidemics, such as complex temperature patterns, population density, wind and transportation patterns, and humidity. We suggest, however, that even the crude nature of the global pattern thus far elucidated is sufficiently compelling that it should be considered for use in global pandemic planning. It is obvious, for example, that new viral variants are most likely to arise where the largest number of susceptibles are exposed to the highest level of viral activity. It would follow, then, that the placement of an efficient array of surveillance centers should be guided by the global distribution of viral activity. It is equally obvious that vaccination strategies appropriate for one level of activity may be less optimal for lower or higher levels of force of infection of the viral agent. Longini et al. [9] demonstrate that the level of coverage needed to produce community protection varies strongly with the local force of viral infection, and earlier models [10,11] demonstrated that factors such as national policy and resource availability interact with the force of infection to determine the locally optimal tactical implementation of influenza control. Every developed country has steadily increased the level of influenza vaccination within its citizenry over time [12], and many are now approaching coverage levels at which discernible effects might be seen. Integration of the global pattern of influenza activity would significantly increase the effectiveness of most of these programs, even at present resource levels and presently enunciated policy provisions.


T.A. Reichert et al. / International Congress Series 1219 (2001) 87–94

References [1] M. Sakamoto-Momiyama, Seasonality in Human Mortality, Univ. Tokyo Press, Tokyo, 1972. [2] T.A. Reichert, A. Sharma, The seasonality of human mortality: the role of influenza, in: A.D.M.E. Osterhaus, et al. (Eds.), Options for the Control of Influenza IV, Elsevier, Amsterdam, 2001(in press). [3] Table 19, 1966 Demographic Yearbook, United Nations, New York, 1968, pp. 361 – 378. [4] Table 32, 1974 Demographic Yearbook, United Nations, New York, 1976, pp. 982 – 999. [5] Table 30, 1985 Demographic Yearbook, United Nations, New York, 1987, pp. 732 – 748. [6] National Geographic Atlas of the World, National Geographic, Washington, 1999, p. 136. [7] W.S. Cleveland, Robust locally weighted regression and smoothing scatterplots, J. Am. Stat. Assoc. 74 (1979) 829 – 836. [8] M. Curwen, Excess winter mortality: a British phenomenon? Health Trends 22 (1990/1991) 169 – 175. [9] I.M. Longini Jr., M.E. Halloran, A. Nizam, et al., Estimation of the efficacy of live, attenuated influenza vaccine from a two-year, multi-center vaccine trial: implications for influenza epidemic control, Vaccine 18 (2000) 1902 – 1909. [10] I.M. Longini, E. Ackerman, L.R. Elveback, An optimization model for influenza A epidemics, Math. Biosci. 38 (1978) 141 – 157. [11] I.M. Longini Jr., J.S. Koopman, M. Haber, G.A. Cotsonis, Statistical inference for infectious diseases: riskspecific household and community transmission parameters, Am. J. Epidemiol. 128 (1988) 845 – 859. [12] F. Ambrosch, D.S. Fedson, Influenza vaccination in 29 countries: an update to 1997, Pharmacoeconomics 16 (Suppl. 1) (1999) 47 – 54.